JobNet: 2D and 3D Visualization for Temporal and Structural Association in High-Performance Computing System

EasyChair Preprint 6941

Abstract

Performance monitoring in a High-Performance Computing (HPC) system is an essential and challenging task. With a large number of system components, coupled with health metrics that need to be reported, visualizing the system's internal structure over time will uncover patterns and enable insights, empowering analysis from monitoring. This paper presents a visualization tool that visualizes the temporal and structural association of HPC system components using a force-directed graph layout algorithm. The visualization contains 2D and 3D representation, supporting a complete analysis of the compute usage, how users and job submission are interconnected throughout the observational interval. Design alternatives for time representation are discussed and depicted in 2D and 3D visualization encodings, with animation and exclusive presentation. The interaction capabilities of the tool assist visual exploration of health metrics and changes in system status over time. The tool's usefulness and effectiveness in the monitoring task are demonstrated by a case study on a real-world HPC dataset.

Keyphrases: 2D and 3D visualization, HPC monitoring, force-directed layout

@booklet{EasyChair:6941,
  author    = {Ngan V.T. Nguyen and Huyen N. Nguyen and Jon Hass and Tommy Dang},
  title     = {JobNet: 2D and 3D Visualization for Temporal and Structural Association in High-Performance Computing System},
  howpublished = {EasyChair Preprint 6941},
  year      = {EasyChair, 2021}}